Understanding slip activity and void initiation in metals using machine learning-based microscopy analysis

Joseph Indeck, David Cereceda, Jason R. Mayeur, Kavan Hazeli

Research output: Contribution to journalArticlepeer-review

Abstract

The use of machine learning techniques to supplement traditional data analysis in mechanics and materials research can improve the understanding of microstructure-property relationships. Identification of key microstructural features or correlation between deformation mechanisms and material response can be discerned that might otherwise have been overlooked. Motivated by the possibilities of gaining additional insight into the process of void nucleation in polycrystalline metals, several machine learning techniques are applied to the analysis of mesoscopic deformation mechanisms as determined by experimental characterization and modeling. Results from crystal plasticity modeling, experimental microstructural analysis, and theoretical models of slip transmission are combined to test a hypothesis regarding fatigue-induced void nucleation. Unsupervised spectral clustering was used with results from crystal plasticity simulations to characterize slip system activity for different crystallographic orientations. The slip system activity as determined by the clustering analysis was then fed into a K-nearest neighbor classifier to quantify the probability of slip transmission across different grain boundaries of interest and analyze grains containing fatigue-induced voids. An unique and unanticipated result from the unsupervised clustering analysis shows that including a group of partially-active slip systems was more appropriate than using the binary classification of active/non-active. Predicted slip activity behavior in a face-centered cubic material was shown to differ significantly from that of a body-centered cubic material due to non-Schmid effects. The outcome of the overall analysis was that grains containing fatigue-induced voids were more likely to be surrounded by grains with orientations that inhibited slip transmission according the Lee-Robertson-Birnbaum (LRB) criteria. Finally, it is demonstrated that smaller datasets using limited simulation results were equally effective at predicting a similar outcome when additional physical descriptors for the slip system activity are used.

Original languageEnglish (US)
Article number142738
JournalMaterials Science and Engineering A
Volume838
DOIs
StatePublished - Mar 24 2022

Keywords

  • Crystal plasticity
  • Fatigue
  • Machine learning
  • Mechanics
  • Slip transmission

ASJC Scopus subject areas

  • Materials Science(all)
  • Condensed Matter Physics
  • Mechanics of Materials
  • Mechanical Engineering

Fingerprint

Dive into the research topics of 'Understanding slip activity and void initiation in metals using machine learning-based microscopy analysis'. Together they form a unique fingerprint.

Cite this